A novel 15-network parcellation from fMRI data in 15 intensively sampled participants

A novel 15-network parcellation of the cerebral cortex, DU15NET, was constructed by estimating networks from functional MRI (fMRI) data in 15 intensively sampled participants (scanned 8-11 times). DU15NET atlases can function and be utilized in group-based studies as a significant update to the previous “Yeo2011” atlas with refinements that align with our current understanding of human cerebral network organization. To access the DU15NET atlases - please click on this link. Upon downloading, you will find three subfolders corresponding to three different spaces: FreeSurfer, HCP, and MNI. Initially computed in FreeSurfer fsaverage6 space, the atlases were further sampled to fsaverage5 and fsaverage space. Additionally, they were projected to HCP fsLR_32k / fsLR_164k surface space and MNI volume space, broadening their utility across various neuroimaging contexts. Here we provide six types of files: DU15NET-Consensus, DU15NET-Consensus-Raw, DU15NET-Prior, DU15NET-Agree53, DU15NET-Agree80, DU15NET-AgreeProb, each described in detail below. For those in search of the primary atlas, please look for DU15NET-Consensus files. For example, the file path ./HCP/fsLR_32k/DU15NET_consensus_fsLR_32k.dscalar.nii represents the DU15NET atlas in HCP fsLR_32k surface space, formatted as CIFTI. If you have any questions about the DU15NET atlases, please contact jingnandu@fas.harvard.edu or randy_buckner@harvard.edu.

References

Du J, DiNicola LM, Angeli PA, Saadon-Grosman N, Sun W, Kaiser S, Ladopoulou J, Xue A, Yeo BTT, Eldaief MC, Buckner RL. Organization of the Human Cerebral Cortex Estimated Within Individuals: Networks, Global Topography, and Function. Accepted for publication. Journal of Neurophysiology. 2024 March.

DU15NET-Consensus: A consensus map represents the assignment at each vertex to the most common network among the 15 individuals collected here. For many purposes, this atlas can function and be utilized in group-based studies as an update to the prior “Yeo2011” atlases with refinements that reflect our current understanding of human cerebral network organization. Given signal dropout near the sinuses and the inner ear, we manually created an uncertain 'network' in the orbitofrontal cortex and adjacent ventrolateral prefrontal cortex (VLPFC), rostral inferior temporal cortex, and the temporal pole, marked as dark grey, we recommend disregarding this 'network' in analyses using this atlas. The DU15NET-Consensus atlas differs from the DU15NET-Prior shown below in that it reflects the idiosyncratic features and spatial registrations of the data from the present participants, including small regions that are absent in the group-averaged estimates. For example, there is a clear representation of the SAL / PMN network in the medial prefrontal cortex.

DU15NET_ColorLUT.txt is a FreeSurfer readable text file specifying how the 15 networks are named, numbered, and colored in Du et al., 2024:

• Index	Network Name	Abbreviation	R	G	B
  0	NONE	NONE	0	0	0	0
  1	Visual Peripheral	VIS-P	170	70	125	0
  2	Cingulo-Opercular	CG-OP	184	89	251	0
  3	Default Network-B	DN-B	205	61	77	0
  4	Somatomotor-B	SMOT-B	27	179	242	0
  5	Auditory	AUD	231	215	165	0
  6	Premotor-Posterior Parietal Rostral	PM-PPr	66	231	206	0
  7	Dorsal Attention-B	dATN-B	98	206	61	0
  8	Somatomotor-A	SMOT-A	73	145	175	0
  9	Language	LANG	11	47	255	0
  10	Frontoparietal Network-B	FPN-B	228	228	0	0
  11	Frontoparietal Network-A	FPN-A	240	147	33	0
  12	Dorsal Attention-A	dATN-A	10	112	33	0
  13	Visual Central	VIS-C	119	17	133	0
  14	Salience / Parietal Memory Network	SAL/PMN	254	188	235	0
  15	Default Network-A	DN-A	100	49	73	0

DU15NET-Prior: The 15-network group prior used in the MS-HBM model is made available as the DU15NET-Prior atlas. The DU15NET-Prior utilizes the HCP data processed via their standard processing pipelines and thus has the benefit of being derived from data collected and registered with a commonly used set of pipelines.

DU15NET-Consensus-Raw: A consensus map represents the assignment at each vertex to the most common network among the 15 individuals collected here, ensuring that no vertex remains unassigned. This map directly reflects the combined outputs from the 15 individual maps collected in this study, , referencing 'DU15NET-Consensus' in both the main text and supplemental figures. The recommended DU15NET-Consensus is subsequently derived based on this map, following a few cleaning procedures outlined in the DU15NET-Consensus section above.

DU15NET-AgreeProb: The agreement atlas represents the probability at each vertex of the consistent assignment of the same network across individuals within the current 15-participant sample.

DU15NET-Agree53: This agreement map is thresholded to leave only vertices with agreement across participants of n ≥ 8 (53%). The agreement atlases may be most useful when the goal is to construct regions of interest that have a high probability of being in one network and not another, in situations where it is appropriate to have regions that cover a portion of the cerebral cortex not the cortex in its entirety.

DU15NET-Agree80: This agreement map is thresholded to leave only vertices with agreement across participants of n ≥ 12 (80%).

Parcellations in Nonlinear MNI152 Volume Space

DU15NET_MNI152_1mm.nii.gz represents the volume version of the DU15NET consensus atlas, consisting of 15 cortical networks, initially derived from DU15NET-Consensus in FreeSurfer fsaverage6 space and then projected into MNI152 space. This transformation is to broaden their applicability across various neuroimaging contexts. The cortical ribbon is defined by putting the FSL MNI152 1mm template through recon-all using FreeSurfer. The same projection process has been applied to all other relevant atlases displayed above.

Information about Downloads

DU15NET, a novel 15-network parcellation from fMRI data in 15 intensively sampled participants. The downloaded folder includes three subfolders corresponding to three different spaces FreeSurfer, MNI, and HCP. In the Freesurfer folder, there are three distinct folders named fsaverage, fsaverage5, and fsaverage6. The fsaverage contains the high-resolution version of the parcellation, while fsaverage6 and fsaverage5 contain lower-resolution versions of the parcellation. Initially computed in FreeSurfer fsaverage6 space, the parcellations were then sampled to fsaverage5 and fsaverage space. Each of these parcels matches up with a network in the 15 network parcellation by Du et al., (2024). It’s important to note that all labels were generated in FreeSurfer version 5.3 and may not fully function optimally in other FreeSurfer versions. Additionally, the atlases were projected to HCP fsLR_32k and MNI space. Additional relevant within-individual parcellation, SNR, and task maps related to DU15NET are available on Balsa (https://balsa.wustl.edu/study/zK166).